unstructured_outlier_mask is not reversely permuted when returned from SPQRUtil.quantize(). Consequently, when comparing the outlier mask and the quantised weights, the positions of the outliers do not correspond.
Is there any reason for this? From what I saw it doesn't affect the functionality of the program whatsoever, but caused me some headaches when studying and debugging the code
unstructured_outlier_mask is not reversely permuted when returned from SPQRUtil.quantize(). Consequently, when comparing the outlier mask and the quantised weights, the positions of the outliers do not correspond.
Is there any reason for this? From what I saw it doesn't affect the functionality of the program whatsoever, but caused me some headaches when studying and debugging the code